

I don’t think so, but LLMs love to write inline multiple page long Python scripts, and since it was taught mostly on sloppy programs in the first place, they are far from optimal.
I’ve seen it overload an admittedly small, but modern database server by mistake.
Point is, on critical hardware, running even nondestructive stuff can become disruptive



I can’t give you anything but “trust me bro” either, but I’m responsible for a couple million in Google Cloud spend, and they’re full out on all European regions, and they tell us so is AWS and Azure.
Right now if you need a gig of RAM, you need to mash the API for days until it finds capacity in most European datacenters. Rep called us and told us we’ll only get what we reserve 2-3 months in advance essentially.
Black Friday is coming up, and there is also all the AI stuff, my bet is that up until now it was mostly a GPU and VRAM crunch, but now with Muse and Instinct it also needs a lot of traditional compute for the harnesses.